E-E-A-T in 2026: what the evidence actually shows

What E-E-A-T actually is

Google's E-E-A-T framework first appeared in their Search Quality Rater Guidelines, a document written for human contractors who manually evaluate search results, not a specification of the ranking algorithm.

In 2013, human raters working for Google began scoring pages and those scores were used to feed quality assessment systems. This was E-A-T (Expertise, Authoritativeness, and Trustworthiness). The connection between rater scores and algorithmic signals is something Google has never explained clearly. It is almost certain the information from these ratings was used to enhance the search algorithm. The first official version was published exclusively for raters in 2013 but then made public in 2015 for ‘greater transparency’ according to Google. However, 2015 was a period of significant tension between Google and the SEO industry and they had spent a lot of time and money trying to remove content designed to manipulate their algorithm using blackhat tactics. One plausible reading is that this public release was at least partly an effort to influence publishers and get them to self-regulate, making Google’s job easier, rather than simply a benevolent desire to increase transparency.

E-E-A-T: experience added

E-A-T later became E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) as part of the updated Google search rater guidelines in December 2022.

The addition of ‘Experience’ was adding a vague concept to an already poorly defined framework. E-E-A-T is clearly not a system. It is a description of what content quality looks like, written for humans.

The evidence gap

No independent controlled study has isolated E-E-A-T as a ranking variable separate from link authority, relevance signals, and domain reputation that predate the concept entirely.

Popular SEO forums and even AI answers give the impression that E-E-A-T is a direct ranking factor, but Google itself has explicitly said this is not the case: “these guidelines are what are used by our search raters to help evaluate the performance of our various search ranking systems, and they don’t directly influence ranking.” There are correlation studies that show high-ranking pages tend to have identifiable authors and external mentions, but correlation is not causation. This evidence amounts to observed correlation with no control for confounding variables.

A confounding factor is a variable that correlates with both the thing being measured and the outcome, making it impossible to isolate which one is actually responsible. In the case of E-E-A-T, the problem is that a page with a named, credentialed author tends to also sit on an established domain with a strong backlink profile, years of indexed content, and high topical authority. Those factors all independently influence ranking. When that page ranks well, there is no way to determine whether the author credential contributed to the ranking, or whether the page would have ranked just as well without it.

To test E-E-A-T as an isolated variable, you would need two otherwise identical pages on otherwise identical domains, differing only in the presence or absence of E-E-A-T signals. That experiment has never been run. There is also a definition problem: E-E-A-T signals are never clearly defined in the first place. Google describes qualities, not measurements. What constitutes sufficient Experience? How much Authoritativeness is enough? The guidelines give raters descriptive criteria, not thresholds. Two raters reading the same page could reasonably reach different conclusions. Without a clear definition of what an E-E-A-T signal actually is, any study claiming to measure its effect is measuring something loosely approximated, not the thing itself.

E-E-A-T has never been tested as a discrete ranking variable. Given the confounding factor problem and the definition problem, it is difficult to see how it could be. Yet it became one of the most discussed frameworks in SEO history. The question worth asking is how that happened.

How an unverified concept becomes industry consensus

E-E-A-T became accepted practice through information cascade, not independent verification: one credible source, repeated through the publication network, produces the appearance of consensus from what is actually a single reflected claim.

AI platforms and Google search are both heavily influenced by consensus corpus. If enough sites are saying the same thing, that thing appears authoritative, and sites that repeat it appear trustworthy. Fabricated or incorrect statements can be amplified by the speed of modern web publishing. The term citogenesis was coined to explain how this happens with errors on Wikipedia and there are a number of documented Wikipedia examples. But this is not just a Wikipedia problem. If a large enough number of sources propagate the same mistruth, an information cascade mechanism is created which feeds the problematic information into Google search and AI models to become the accepted version of events, regardless of whether the original claim was ever verified.

E-E-A-T is a clean example of this mechanism in action. Google published a document describing what quality content looks like for human raters. SEO publications covered it as a ranking system. Agencies built services around those coverages and had a direct commercial interest in the concept gaining traction. AI systems trained and retrieved from that saturated corpus now return E-E-A-T as established fact when asked about ranking factors. As more people read those AI summaries of E-E-A-T they in turn post about it, creating more content for AI platforms to retrieve. The loop closes without anyone checking whether the original claim was sound. The original claim, that these guidelines inform quality assessment without directly influencing ranking, got lost somewhere between the first agency blog post and the ten thousandth. Nobody ever went back to test it.

The schema markup claim followed the same pattern, but in that case controlled experiments eventually caught up with it. SearchVIU tested five major AI systems and found that information placed only in JSON-LD schema, invisible in the page content, was extracted by zero out of five platforms. Ahrefs tracked 1,885 pages before and after schema implementation and found citations barely moved, though it is worth noting the study only measured pages already receiving significant AI citations. The 2026 AI Citation Visibility Study, which audited 50 crypto protocols across ChatGPT, Perplexity, and Google AI Overviews, found schema scores did not predict citation visibility. Years of confident industry consensus, built through the same information cascade mechanism, turned out to be wrong. The evidence was always obtainable. But nobody had bothered to obtain it.

YMYL: slightly more observable, same fundamental problem

The YMYL (Your Money or Your Life) category has more observable evidence than E-E-A-T through the pattern produced by the 2018 Medic Update, but the causal mechanism remains unconfirmed, and the category boundaries have expanded to the point of near-meaninglessness.

Google introduced the term in its Search Quality Rater Guidelines to describe content where a low-quality page could cause real harm: wrong medical advice, bad financial guidance, or misleading legal information. The definition was formalised in the 2018 revision of the SQRG. The guidelines are used by quality raters to assess content quality, but not to directly set rankings. Danny Sullivan, Google Search Liaison, has said publicly that the rater programme informs how Google evaluates its own systems, not individual pages. That is an important distinction.

In August 2018, Google rolled out a broad core update in which many health and finance sites were adversely affected. Barry Schwartz named it the Medic Update. But the pattern does not confirm the mechanism.

Sites that dropped had, in most documented cases, more than one problem simultaneously: anonymous authorship, poor content, no verifiable expertise signals, poor external authority and third-party coverage. Isolating YMYL category treatment as the cause is not possible from external observation. Google confirmed the update as a broad core algorithm change and said nothing more specific about YMYL.

The boundaries of the YMYL definition have expanded with each revision. Today, most content that could influence a financial or health decision qualifies as YMYL. Crypto content likely sits comfortably inside that boundary.

There does seem to be a pattern consistent with stricter treatment of YMYL content, though there is no confirmed mechanism. The uncertainty does not change the practical decision. Anonymous, unattributed content making implicit financial claims is unlikely to rank well, so operating as if YMYL standards apply is a rational risk management decision.

Why corpus consensus can fail as a quality signal

AI retrieval systems and search engines weight breadth of representation in the source corpus, which means widely repeated claims appear authoritative regardless of whether the original claim was verified before it propagated.

A claim present in fifty sources outscores a careful rebuttal present in two, even when the rebuttal is better evidenced. The retrieval stage does not assess the quality of the original claim. It measures how consistently that claim appears across the training corpus and, in retrieval-augmented systems, across the indexed sources pulled at query time. Repetition functions as a proxy for authority, whether the original claim was sound or not. AI systems have no reliable way to distinguish between the two before surfacing an answer.

Search ranking has a related problem. Link-based authority and topical signals reward content that matches existing patterns. A page covering a topic in the way that existing high-ranking pages cover it can earn relevance signals from that alignment. This is why compiled content can rank: it reproduces the structural consensus of the source corpus and receives the topical signals that consensus generates. It ranks until something less compiled or more authoritative displaces it, which requires a page with original evidence, a named claim, or a source willing to vouch for it editorially. Until that page exists, the compiled version holds position by default.

The schema markup story is the clearest available example of corpus consensus failure with evidence attached. Across the web, schema markup was widely described as improving AI citation visibility. The claim propagated through SEO publications, agency blogs, and practitioner guides until it became perceived wisdom. The 2026 AI Citation Visibility Study, which tested 50 crypto protocols across ChatGPT, Perplexity, and Google AI Overviews, found no support for it. Controlled experiments from SearchVIU and Ahrefs, published independently, produced direct evidence that the mechanism was wrong.

Using E-E-A-T as a content framework rather than an algorithmic target

The E-E-A-T framework describes properties of credible, citable content with reasonable accuracy, and those properties are worth building regardless of whether Google has a discrete E-E-A-T signal in its ranking algorithm. When creating content, crypto projects would be wise to concentrate on producing useful content that is worth citing, rather than working to an E-E-A-T compliance checklist.

Experience

Experience in this context means first-person observations that demonstrate actual use and familiarity: specific details only someone who has worked in the space would notice. A paragraph that includes a specific insider observation or opinion is more useful than another explainer that simply restates a general mechanism. The mechanism is likely already available everywhere. The observation is unique.

Insider observations are difficult to generate from a SERP analysis using automated tools. Including genuine experience when writing is an integral part of creating citable content that is extractable from an AI-generated answer rather than absorbed into one.

Expertise

Expertise in content terms is named claims with sources. A specific figure attributed to a specific study. A position stated and then supported with evidence the reader can follow back to its origin. This is the difference between content that has a citation moment and content that does not.

Compiled content reproduces the structural consensus of existing sources without adding original thinking. It may be accurate. It may be well-written. It has no citation moment, because there is no sentence in it that a writer elsewhere would reach for to support a claim of their own. Expertise, as a content property, is what creates that sentence. It does not require an academic credential. It requires a claim specific enough to be useful and a source traceable enough to be trusted.

Authoritativeness

Authoritativeness is an external property. It exists in what third-party sites say about you, not in what you say about yourself. Coverage in third-party publications, citations from independent researchers, a named author with a verifiable presence outside the site in question: these are the signals that establish authority in practice.

The 2026 AI Citation Visibility Study found third-party source coverage to be the dominant factor in AI citation visibility across the platforms audited. Protocols with the deepest independent coverage were cited most frequently, irrespective of their own content quality scores. An anonymous page, however technically accurate, contributes no named entity to the knowledge graph associations an AI model draws on. A named author with external presence does.

Trustworthiness

Trustworthiness in a content context is transparency: about methodology, about limitations, about what the evidence does and does not show. The 2026 AI Citation Visibility Study’s limitations section is a practical example. It states explicitly that the sample covered 50 protocols, that citation rates were measured at a single point in time, and that the correlation findings do not establish causation. Those disclosures do not weaken the study. They make its conclusions more defensible, because a reader who reaches for it to support their own argument can do so without worrying that a caveat will undermine them later.

Most crypto project teams that have been publishing for twelve months and seen no traction have the same underlying problem. The posts exist. There is just nothing in them a reader would reach for to support a claim of their own. The writing may be competent. The sourcing is thin, the authorship is anonymous, and the methodology, where it exists at all, is never stated. Whether that is a trustworthiness failure in Google’s terms is unverifiable. It has no citation moment.

Frequently asked questions

• Is E-E-A-T a real ranking factor?

  E-E-A-T is not a ranking factor in the sense of a measurable score Google applies to pages. Google has confirmed this directly. It is a framework used in the Search Quality Rater Guidelines to describe what high-quality content looks like, and quality raters manually assess content against it. The properties it describes, named authorship, verifiable expertise, external coverage, transparent sourcing, do correlate with ranking performance, because those same properties also drive link acquisition, entity recognition, and topical authority signals.

• Does Google actually use E-E-A-T scores?

  There is no E-E-A-T score. Google has said so explicitly. Quality raters use the framework to evaluate whether Google’s systems are surfacing good results, not to rank individual pages. The practical implication is that optimising for a score that does not exist in the algorithm is the wrong frame. The right frame is that the properties E-E-A-T describes, such as original experience, sourced expertise, independent authority signals, and honest methodology, help make content linkable, citable, and extractable by AI retrieval systems. Building those properties is worth doing regardless of whether Google applies an E-E-A-T score.

• Will E-E-A-T get my crypto protocol cited in AI answers?

  Not directly, and the framing matters here. AI citation visibility is driven primarily by third-party coverage: how many independent sources reference your protocol, how clearly your entity is defined across those sources, and whether your own content contains specific, extractable claims. Those are the same properties the authoritativeness and expertise components of E-E-A-T describe. The 2026 AI Citation Visibility Study found that protocols with the deepest independent third-party coverage were cited most frequently across ChatGPT, Perplexity, and Google AI Overviews, irrespective of schema quality or on-site content scores. E-E-A-T signals and AI citation visibility point at the same underlying problem.

• E-E-A-T vs schema markup: which matters more?

  For AI citation visibility, third-party coverage matters more than either. For search ranking, the E-E-A-T properties have more practical impact than schema alone: named authorship, sourced claims, and external authority signals all matter. Schema’s durable value is narrow. Organisation schema helps establish your entity in Google’s knowledge graph, and Article schema aids content parsing. It does not drive AI citations. The 2026 AI Citation Visibility Study found schema quality scores did not predict citation rates across 50 audited crypto protocols. Aave, which has no schema, ranked second overall. Treat schema as a declaration of what your content already establishes, not as a substitute for establishing it.